OSOcean ScienceOSOcean Sci.1812-0792Copernicus PublicationsGöttingen, Germany10.5194/os-12-51-2016Investigation of model capability in capturing vertical hydrodynamic coastal processes: a case study in the north Adriatic SeaMcKiverW. J.SanninoG.https://orcid.org/0000-0002-3985-9432BragaF.https://orcid.org/0000-0002-4131-9080BellafioreD.debora.bellafiore@ismar.cnr.itISMAR-CNR, Arsenale – Tesa 104, Castello 2737/F, 30122 Venice, ItalyENEA Centro Ricerche Casaccia, Via Anguillarese 301, 00123 Rome, ItalyD. Bellafiore (debora.bellafiore@ismar.cnr.it)15January2016121516926June20153August20154December201517December2015This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://os.copernicus.org/articles/12/51/2016/os-12-51-2016.htmlThe full text article is available as a PDF file from https://os.copernicus.org/articles/12/51/2016/os-12-51-2016.pdf
In this work we consider a numerical study of hydrodynamics in the
coastal zone using two different models, SHYFEM (shallow water hydrodynamic finite element model) and MITgcm (Massachusetts Institute of Technology general circulation model), to
assess their capability to capture the main processes. We focus on
the north Adriatic Sea during a strong dense water event that
occurred at the beginning of 2012. This serves as an interesting
test case to examine both the models strengths and weaknesses, while
giving an opportunity to understand how these events affect coastal
processes, like upwelling and downwelling, and how they interact
with estuarine dynamics. Using the models we examine the impact of
setup, surface and lateral boundary treatment, resolution and mixing
schemes, as well as assessing the importance of nonhydrostatic
dynamics in coastal processes. Both models are able to capture the
dense water event, though each displays biases in different
regions. The models show large differences in the reproduction of
surface patterns, identifying the choice of suitable bulk formulas
as a central point for the correct simulation of the thermohaline
structure of the coastal zone. Moreover, the different approaches in
treating lateral freshwater sources affect the vertical coastal
stratification. The results indicate the importance of having high
horizontal resolution in the coastal zone, specifically in close
proximity to river inputs, in order to reproduce the effect of the
complex coastal morphology on the hydrodynamics. A lower resolution
offshore is acceptable for the reproduction of the dense water
event, even if specific vortical structures are missed. Finally, it
is found that nonhydrostatic processes are of little importance for
the reproduction of dense water formation in the shelf of the north
Adriatic Sea.
Introduction
Coastal hydrodynamic processes play an important role in ocean
dynamics. Being at the interface between land and sea, they are
strongly influenced by the input of freshwater through river
discharge, tides, topographic features, as well as human activities
and are affected, at the surface, by the winds and by heat and water
fluxes.
The hydrodynamics typically observed in coastal areas involve processes
interacting on a wide range of spatial and temporal scales, as well as
slowly and rapidly varying features . The scale
interaction seen in the coastal zone is driven by a number of local
(wind, sources of freshwater) and large-scale (pressure, surface heat
and mass fluxes) forcings. Along the coast, the surface wind affects
the dynamics of freshwater from river inputs with different wind regimes
causing the buoyant flow to narrow or thicken, leading to
increased upwelling or downwelling .
Moreover, the interaction with coastal water bodies leads to the
identification of “regions of freshwater influence” (ROFI hereafter)
and interaction zones in the proximity of lagoons and transitional
areas . The presence of complex coastal
morphologies, embayments, promontories, and sudden bathymetric changes
can interact with coastal currents producing small-scale features (filaments) with specific temporal and spatial variation
. Islands, as well, can
enhance small-scale features and are characterized by vertical
movements during specific tide and wind conditions
. In fact, the majority of horizontal
structures observed in the coastal zone are characterized by Rossby
and Richardson numbers of around 1 (submesoscale), representing
areas of frontogenesis where vertical fluxes and buoyancy are enhanced
. In such a complex environment, sudden changes in
the forcings can trigger strong hydrodynamic events, such as the
formation of dense water (DW), wind-driven upwelling, and peak river
floods.
Modeling the coastal zone and the specific hydrodynamic processes
occurring there is challenging due to the number of spatial scales
involved and the complex morphologies . In
particular, these processes can produce strong vertical motions, which
are difficult to model, requiring high resolution and an accurate
representation of the underlying physics, perhaps even requiring the
inclusion of nonhydrostatic processes. Improving modeling skills for
reproduction of coastal processes is a balance between trying to
capture the full range of physical processes involved (turbulence,
mixing, non-hydrostatic vertical motion), while at the same time
introducing suitable numerical approaches for efficient simulation of
the processes. Modeling tools, with appropriate horizontal and
vertical discretization, are needed (finite difference – finite
volumes – finite elements; structured – unstructured grids). Also the
choice in numerical parameterization schemes, particularly concerning
vertical mixing, play a central role .
When modeling vertical processes, one issue to consider is whether
nonhydrostatic processes are important for reproducing them. Several
studies investigated this issue: studied the
effect on submesoscale processes, stating the difficulty in
identifying specific vertical features connected with nonhydrostatic
process modeling. A major effect of the choice of resolution in
pattern reproduction is stressed. found that
nonhydrostatic effects do not play a major role in coastal upwelling,
but, interestingly, they identify their impact on the horizontal
patterns (enhanced meandering). stressed how the
nonhydrostatic processes can affect the energy transfer between scales
and they pointed out the need to investigate the possible role of
nonhydrostatic processes in quantifying the modulation of scale
interaction (on the horizontal) along the coast.
The Adriatic Sea is an example of a water body that is strongly linked
to its coastal system, being a semi-enclosed basin with a particular
topography, i.e., having a very shallow northern area becoming deeper
towards the south, and a large number of freshwater sources
. This makes it prone to DW
events, when cold north-easterly winter winds induce water sinking in
the shallow northern Adriatic . These
extreme DW events have many complex influences and thus are
particularly challenging to understand and model, though their impact
on the general circulation has made them an important topic of
research . Here we focus on one particularly strong DW
formation event that occurred in the beginning of 2012. The extreme intensity
of this DW event motivated many studies, with a large collection of in situ
data, providing insights on the hydrodynamic features that occurred.
Therefore, this case serves as an interesting test to assess the models,
allowing us to compare our results with previous study efforts
, while having an opportunity to complement the
understanding of how these events affect coastal hydrodynamic
processes and, in particular, probing into what are the most suitable
modeling strategies to reproduce them.
There are still many aspects of coastal dynamics that are not well understood
and there are limits to the information garnered from in situ observations
and measurement campaigns. Much about the dynamics of these processes must be
studied through the use of numerical models. With this in mind, our approach
here is to use two very different numerical models, SHYFEM (shallow water hydrodynamic finite element model) and MITgcm (Massachusetts Institute of Technology general circulation model), in order to compare their strengths
and weaknesses in representing these processes. In particular we assess their
ability to capture the DW event, its formation and propagation, as well as
associated coastal upwelling and downwelling. In addition to a comparison
between the two different models we also compare two simulations, one
imposing hydrostatic balance, the other fully nonhydrostatic, in order to
determine what impact nonhydrostatic processes have in regional coastal
processes and DW phenomenon.
In Sect. we describe the models used and simulation
setup, as well as providing a list of observational data used for
comparison with the models. In Sect. we present the
results, beginning in Sect. with a validation of the
models against observational data. In Sect. we take
a broad look at how the models represent the coastal dynamics, namely
the DW formation and propagation, followed by an analysis of the
coastal upwelling (Sect. ), and the impact of estuarine
dynamics (Sect. ). We discuss the results in
Sect. and draw our conclusions in Sect. .
Methods
In this study we use two different three-dimensional (3-D)
hydrodynamic models, SHYFEM and MITgcm. Both are designed for
oceanographic studies and both have been previously applied in the
open sea and in the coastal area of the Adriatic Sea
.
SHYFEM
The SHYFEM model has a finite
element grid covering the Adriatic Sea (excluding the lagoons) consisting of
23 657 nodes, 43 768 elements and 59z layers in the vertical, with
different thicknesses up to a maximum depth of 1280 m. The bathymetry
used is a merge of data from the NURC (NATO Undersea Research Centre) data set provided within the
ADRIA
02 framework and field campaigns done by ISMAR-CNR (Institute of Marine Science – National Research Council) within the last 15 years in
the area in front of the Venice Lagoon. Water levels are set at the mean sea
level as an initial condition and are then adjusted to the computed values.
3-D velocity values are initially set to zero. The main open boundary is
located at the Otranto Strait. 3-D temperature and salinity and tidal water
level time series force the open-boundary section. At the lateral open
boundaries, corresponding to river inflows, discharge time series are
imposed. Bottom stress is applied using a constant bottom friction
coefficient (0.0025). We use a TVD (total variation diminishing) scheme for both the horizontal and
vertical advection in the transport and diffusion equation for scalars, with
constant diffusivity (0.2m2s-1). Horizontal advection of
momentum is discretized by an upwind scheme and horizontal eddy viscosity is
computed by the Smagorinsky's formulation. For the computation of the
vertical viscosities and diffusivities, a k–ϵ turbulence scheme
is used, adapted from the GOTM (General Ocean Turbulence Model) model described in
.
On the surface, a constant value for the wind drag coefficient is used
(0.0025). To reproduce the surface heat fluxes, shortwave radiation
from the atmospheric model is imposed, whereas the long-wave radiation
is computed according to the formula. Bulk
formulas are computed considering the sea surface temperature, the
winds at 10m height, the dry air temperature and the air
pressure at 2m, and the relative humidity as inputs. The
latent heat flux and the sensible heat flux are computed according to
the bulk formula. Cloud cover is taken from the
atmospheric model.
MITgcm
The MITgcm solves both the hydrostatic and nonhydrostatic
Navier–Stokes equations under the Boussinesq approximation for an
incompressible fluid with a spatial finite-volume discretization on
a curvilinear computational grid. The model formulation, which
includes implicit free surface and partial step topography, is
described in detail by . The model domain, that covers the entire
Adriatic Sea, is discretized by a non-uniform curvilinear orthogonal
grid of 432×1296 points. The model has 100 vertical z
levels with a thickness of 1m in the upper 23m
gradually increasing to a maximum of 17m for the remaining
64 levels. The bathymetry used by MITgcm is provided by the National Group
of Operational Oceanography (GNOO; http://gnoo.bo.ingv.it/bathymetry/).
As in and , an
implicit linear formulation of the free surface is used. The model
uses constant horizontal eddy coefficients for momentum (viscosity:
10m2s-1), temperature, and salinity (diffusivity:
2m2s-1). Vertical eddy viscosity and diffusivity
coefficients are computed in the MITgcm using the turbulence closure
model developed by for the atmosphere and
adapted for the oceanic case by .
The river runoff is considered explicitly and modeled as a lateral
open-boundary condition. As in , the rivers are
included by introducing small channels in the bathymetry that simulate
the river bed close to the coast. Velocity is imposed at the upstream
end of each channel, with the prescribed discharge rate being obtained by
multiplying the velocity by the cross sectional area of the channel.
No flux conditions for either momentum or tracers and no slip
conditions for momentum are imposed at the solid boundaries. Bottom
drag is expressed as a quadratic function of the mean flow in the
bottom layer: the (dimensionless) quadratic drag coefficient is set
equal to 0.002.
The net transport through the southern open boundary is corrected
during run-time at each time step to balance the effects of river
discharge and of the evaporation minus precipitation budget on the
surface level. This solution prevents any unrealistic drift in the sea
surface elevation. Tides are imposed as a barotropic velocity at the
southern boundary.
At the surface, the wind drag coefficient is computed following the
default MITgcm formulation:
Cd=0.0027U10+0.000142+0.0000764U10,
where U10 is the wind speed at 10 m. The treatment of
surface heat forcing is done with the same bulk formula used in
SHYFEM, except for the sensible and latent heat fluxes, where the
formulation proposed in is
used.
Simulation setup
Two numerical experiments were carried out. The first experiment is concerned
with how the two models compare during the DW event of 2012. In this
experiment both models are implemented with hydrostatic balance. Both model
simulations begin in December 2011 and are run until the end of April 2012.
This period covers the DW event in the beginning of 2012. The time steps used
for SHYFEM and MITgcm are 20 and 10 s, respectively. Output fields and
diagnostics are produced every three simulated hours. Surface forcings (wind
speed and direction, air temperature, relative humidity, and cloud cover) are
provided by means of hourly meteorological forecasts from the
MOLOCH (MOdello LOCale in H coordinates)
model . The MOLOCH model is a
non-hydrostatic atmospheric model running on a horizontal grid with 2.3 km
resolution and 54 vertical layers, developed and run at the
ISAC (Institute of Atmospheric Sciences and Climate – National Research Council)-CNR, Bologna, Italy
. The atmospheric model allows for the investigation
of the effects of local and highly variable atmospheric processes in the
coastal area, due to its high resolution. Temperature and salinity are
initialized, interpolating 3-D values on the two grids, and forced at the
open boundary at the Otranto Strait, from AFS (Adriatic Forecasting System)
data. AFS data are forecasts providing daily mean 3-D fields on a sigma level
system with 2km horizontal resolution. Tidal water level and surge
data are provided from the OTIS (OSU tidal inversion software) tidal model and AFS sea-surface
height data. River inputs have been included for the Po, Adige, Brenta,
Livenza, Piave, Tagliamento, and Isonzo rivers. The Po River discharge is
provided by ARPA (Agenzia regionale per la prevenzione e protezione ambientale) Emilia Romagna
(ARPA-SIMC (Servizio Idro-Meteo-Clima)), daily values. The Tagliamento and Isonzo rivers discharge are
provided by Regione Friuli Venezia Giulia (Servizio Idrografico) with
a frequency of 30 min and are measured by two tide gauges in front of the
river mouths. The period chosen for the present run is not covered by
measured discharge data for the other rivers; therefore, climatological
values computed on a large daily data set covering the period 2005–2010 are
used. All the river boundaries are forced with measured water temperature
time series from the year 2007, collected on the Tagliamento, except the
Isonzo River that uses its own measured time series, available for the same
year. Where data are missing in the Tagliamento and Isonzo measured time
series, gaps are filled with climatological data.
In the second experiment the nonhydrostatic version of the MITgcm
model is run, again over the same time period, to assess the
importance of nonhydrostatic processes.
As the two models have different grids their resolutions are
considerably different. In Fig. we show maps of the
difference in resolution of the two models, with red and blue
indicating where MITgcm is more or less resolved than SHYFEM,
respectively. As can be seen overall the MITgcm has higher
resolution. Only in coastal regions do the models have comparable
resolution, with the blue regions in the right panel indicating where
SHYFEM is more resolved.
Maps of the difference in grid resolution (in km) between
MITgcm and SHYFEM (SHY minus MIT) in the entire Adriatic (top left),
and a close up of the north-eastern coastal area, the Gulf of
Trieste (bottom left). Red indicates where MITgcm is higher resolved
than SHYFEM while blue indicates the reverse. The two grids, SHYFEM
(top right) and MITgcm (bottom right) are shown for the area of the
Gulf of Trieste.
Observational data
In order to validate the model simulations, a number of observational
data sets are used. Figure shows their location.
CTD (conductivity, temperature, and depth) transects of temperature and salinity are provided from
a cruise with the R/V DallaPorta, along the Senigallia transect (Fig. ), where temperature and salinity profiles
were acquired with a SeaBird Electronics SBE 911-plus CTD, on the
27 March 2012. These sets of data are part of a larger data set,
collected in the bimonthly monitoring activity along that transect.
Sea surface temperature (SST) from satellite data obtained
using Moderate Resolution Imaging Spectroradiometer (MODIS). MODIS
is a key instrument aboard the NASA Terra and Aqua satellites, which
acquires measurements in 36 spectral bands. It can provide a wide
range of atmospheric, land, and oceanic products: specifically for
the ocean, MODIS SST is retrieved from radiometric measurements at
11 and 4µm wavelengths with 1km of
spatial resolution. We selected the MODIS-Aqua SST for the Adriatic
Sea, acquired during daytime on the 26 January, 5 and 16
February 2012, and available on the OceanColor web page of the
Goddard Space Flight Center of NASA
(http://oceancolor.gsfc.nasa.gov/). Since the SST products
were highly affected by clouds, they have been adequately cloud
masked, with the MODIS atmosphere products
(http://modis-atmos.gsfc.nasa.gov). To prevent the loss of
river plume information in the area close to the coast, a quality
mask was not applied to the SST, because the effects on the SST
fields were negligible as tested by . Finally, SST
fields were remapped to geographical lat–long coordinates.
In the Gulf of Trieste, time series of surface
(2.5m) temperature and salinity and bottom
(22.5m) temperature from the Vida buoy (location
45∘32′55.68′ N,
13∘33′1.89′′ E; Fig. )
are used to validate the models output and analyze the thermohaline
variation in the simulated period. Also surface (2m) and
bottom (12m) temperature, salinity, and density anomaly
from sensors installed at the CNR Platform Acqua Alta (location
45∘18′49.8′′ N, 12∘30′31.8′′ E; called
hereafter AA Platform; Fig. ) are available for model
validation, for the time window 1 December 2011 to the 31
March 2012.
Map showing location of data sources used for validating the
model: CTD data (red dots), Acqua Alta CNR Platform (yellow dot), and
Vida buoy (green dot). Purple dots show the location of river
inputs.
Results
The backdrop for our simulations is the extreme DW outbreak that
occurred during the winter of 2012. In the Adriatic Sea during the
period January–February, there was an unprecedented generation of DW
with record breaking density anomalies of above 5kgm-3
relative to a value of 1025kgm-3. The event took place after a particularly
warm and dry year, resulting in a reduction of coastal freshwater
supply, in the backdrop of an already long-term trend in increasing
salinity. The event was then triggered by an extended period of cold
weather with strong Bora winds that lasted for about 3 weeks in
the coastal eastern Adriatic region, between the 25 January and the 14
February 2012 . In what follows we will show
how the two different models capture various aspects of this
phenomenon, beginning first with a comparison with the measurements
available for this period and then showing how the models reproduce
a number of the specific processes affecting the coastal zone, namely
the timing of the DW outbreak, its formation and evolution, coastal
upwelling and riverine processes. We also look at a comparison of the
hydrostatic and nonhydrostatic simulations.
Model validation
Here we provide an assessment of how well the two hydrostatic models
do in reproducing the hydrodynamics of the coastal zone, as seen in
the observations. In Fig. the time series of surface and
bottom temperature as well as surface salinity from the Vida buoy is
shown. After the strong cold Bora wind event, the surface waters lose
heat leading to DW sinking. The monotonic decrease of temperature
continues until the beginning of February when the sudden loss of
surface heat produces an abrupt drop in temperature from about 10 to
6∘C (2–3 February), before climbing up to
8∘C a couple of days later (Fig. a). On the
bottom, temperatures reach even lower values (5∘C around
the 15 February), suggesting an injection of cold water down from the
surface from the areas in the vicinity during the Bora event. In the
lead up to the event, up to the end of January, both models reproduce
well the surface and bottom temperatures. Also they both capture the
onset of the event, registering the starting moment of the cold water
sinking at the beginning of February. However both models overestimate
the minimum temperature values reached during the event. In the case
of SHYFEM, the surface values are well represented; however, bottom
temperatures are overestimated, with their values being close to those
of the surface, indicating that the model mixes the water column too
quickly and does not show any significant unstable stratification due
to the DW sinking. MITgcm reproduces the surface temperature before
the event, with just a small underestimation in the first simulated
month. However, bottom temperature is closer to the observations during
the event, and the greater difference in the temperature values
between surface and bottom, indicates that MITgcm has a more unstable
stratification and less mixing than SHYFEM during the DW event.
Time series of (a) surface (left panel) and bottom (right
panel) temperature and (b) surface salinity for SHYFEM (blue),
MITgcm (red) and the Vida buoy observations (black).
From the statistical analysis of the whole temperature time series,
which is shown in Table , it is evident that the two
models well reproduce measurements, with biases always lower than
0.2∘C. Correlation is higher than 0.96 for SHYFEM
surface and bottom temperature data, while slightly lower for
MITgcm. The two models show higher errors in reproducing the
time series variability, as expressed by root mean square error (RMSE) values of around
1∘C.
Statistical analysis of simulated water temperature and salinity time series computed at the Vida buoy. Analyses provided are the difference between mean
of observations and simulations (Bias), the root mean square error (RMSE), and the correlation.
LocationVariableSHYFEMMITgcmBiasRMSECorrelationBiasRMSECorrelationTemperature (2 m) [∘C]-0.140.830.960.191.110.88Vida buoyTemperature (22 m) [∘C]0.111.040.98-0.041.120.94Salinity (2 m) [psu]0.300.350.840.310.370.73
The Vida buoy also shows a general increase of surface salinity during
the first months of 2012 (Fig. b), probably connected with
the low discharge of freshwater characterizing the whole north
Adriatic Sea (NAS) in that period, in particular the closest river
Isonzo . Both models overestimate the
surface salinity during the whole period, as shown in
Table (bias around 0.3psu for both models),
with SHYFEM showing a lower variability compared to MITgcm
results. Correlation for surface salinity is higher for SHYFEM
(0.84) than for MITgcm (0.73), but high enough for both models to
state that the reproduction capability of the haline temporal
evolution is matched (Table ).
Timeseries of surface (solid) and bottom (dashed) (a)
density
anomaly, (b) temperature, and (c) salinity for
SHYFEM (blue), MITgcm (red), and the CNR Platform observations (black). Note
there is a gap in the data for the bottom observations between the 8 and 25
January.
The AA Platform surface (-2m) and bottom (-12m)
data for density anomaly, temperature, and salinity (Fig. )
have a similar trend as that seen in the Vida buoy time series, with
the density anomaly peak reached at the end of the first week of
February 2012. The measurements reveal the stable stratification, just
before the DW formation event (density anomaly difference, between
surface and bottom, around 1kgm-3, even if the water
column is thermally unstably stratified), the passing of the well-mixed DW (until the 22 March), and the subsequent
re-stratification. SHYFEM does match the stable density stratification
before the event, just for a few days, even if it has a slightly more
homogeneous water column, compared with measurements
(Fig. a). The DW signal is registered by SHYFEM perfectly
matching the density anomaly values in the month of February. The
major discrepancy is in the reproduction of the surface density
anomaly after the event when a mass of lighter water is measured on
the surface. MITgcm, as well, matches the general trend, with density
anomalies closer to the measured ones, before the event, compared with
SHYFEM, but overestimating the values during the event, in February.
Temperature trends are well matched by both models: Table
shows correlation values, for SHYFEM, of 0.88 and 0.97, for
surface and bottom temperature, respectively. Also MITgcm shows high
correlation values for temperature, even if slightly lower than SHYFEM
(0.77 and 0.86 on surface and bottom, respectively). SHYFEM
better reproduces the temperature variability before the event and has
a better match with observations in the post-event period, compared
with MITgcm. Due to the lack of measurements of bottom temperature in
the period just before the event, it is not possible to state whether
the unstable thermal stratification, reproduced by SHYFEM or the well-mixed thermal structure, simulated by MITgcm at the AA platform,
represents the real process. The salinity time series indicates that
the density anomaly discrepancy in the models is due to the freshwater
dynamics, specifically the lack of direct measurements to impose as
input for the studied period, as was the case at the Vida buoy. In
fact higher salinity biases are registered by both models on the
surface (1.18 and 1.07psu for SHYFEM and MITgcm,
respectively) while a better match is seen at the bottom
(Table ). Clearly river inputs that provide an incorrect
amount of freshwater discharge, can directly affect the salinity
variation (seen in Table , with very low correlation
values for the two models). The salinity mismatch also affects the
surface density anomaly (Table ).
Statistical analysis of simulated water temperature and salinity time series computed at the Acqua Alta CNR Platform. Analyses provided are the difference between mean
of observations and simulations (Bias), the root mean square error (RMSE), and the correlation.
LocationVariableSHYFEMMITgcmBiasRMSECorrelationBiasRMSECorrelationTemperature (2 m) [∘C]0.891.420.881.351.970.77Temperature (12 m) [∘C]0.270.870.970.401.330.86AA PlatformSalinity (2 m) [psu]1.181.700.111.071.460.63Salinity (12 m) [psu]0.420.610.370.270.380.84Density Anomaly (2 m) [kgm-3]0.791.180.530.610.900.79Density Anomaly (12 m) [kgm-3]0.350.560.910.140.370.90
In Fig. we show comparisons of model temperature and salinity
with three CTD profiles from the Senigallia transect, moving away from the
coast the profiles are indicated with dash-dot, dashed and solid lines
respectively (location indicated in Fig. ) for the 27 March 2012.
The DW signal, produced at the beginning of February in the northern end of
the basin, flowed along the Italian shelf and can be detected at the bottom
around 20km offshore, from CTD profiles (dashed profiles). SHYFEM
reproduces the salinity profiles with a general underestimation of
0.5psu. Generally MITgcm overestimates coastal salinity, while the
more offshore profiles show similar differences from measurements for both
models. Coastal haline stratification could be missed as a consequence of the
Po River plume mismatch, which is discussed below. SHYFEM shows quite a clear
underestimation of surface temperature, particularly offshore, with a bias of
about -1∘C. However it matches better the data along the entire
water column. MITgcm overestimates the surface temperature by
2∘C while it underestimates the bottom values by
1∘C.
Senigallia transect: profiles of salinity (left) and
temperature (right) for three CTD profiles along the transect for the 27
March 2012 (big red dots shown in Fig. 2). The inset shows comparison
between observations (black), SHYFEM (blue) and MITgcm (red) for the
innermost, shallower profile (dash-dot), one in the center of the
transect (dashed), and the outermost, deeper profile (solid).
MODIS SST images (left column) and bias maps showing the
difference between model SST from MODIS satellite observations for
SHYFEM (center column) and MITgcm (right column), for times before,
during, and just after the dense water event, namely 26 January (top
row), 5 February (center row), and 16 February 2012 (bottom row). Units
are [∘C].
In Fig. we show maps of the SST
from MODIS satellite observations and model minus satellite
differences for both models, for three different times: before, during,
and just after the DW period (26 January, 5 and 16 February). The
comparison shows generally common behavior for both models, with
differences in small-scale features. The 26 January satellite SST
reveals a bulk of cold water, as is typical of the winter season,
flowing out from the Po River that produces a clearly identifiable
strip of coastal cold waters along the Italian littoral, just south of
the river mouths. The cold discharge from the northern rivers is
detected and the whole coast is characterized by a SST lower than
6∘C. SHYFEM and MITgcm overestimates the surface
temperature of these waters, by around 2∘C. MITgcm tends
to slightly overestimate (between 0.5 and 1 ∘C) the whole area of
NAS, except for a small area in the Gulf of Trieste, where also SHYFEM
displays a slight underestimation (-0.5∘C). If results
are considered in a basin-wide perspective, SHYFEM tends to
underestimate SST in the deepest areas offshore, located in the center
of the Adriatic Sea, while the biggest errors for MITgcm, are detected
along the Italian littoral in the middle Adriatic Sea, with a strip of
coastal waters underestimated by more than 2∘C
(Fig. top). Better performances of both models can be seen
in the comparison with satellite images for the two dates during and
just after the DW formation event. Still there is an overestimation of
temperatures in the narrow strip in the proximity of rivers but the overall bias
is reduced, in the range [1,+1]∘C. A major discrepancy is seen in the
reproduction of a cold structure just offshore of the Croatia
littoral. Both models overestimate SST there and this suggests that
a specific process is not reproduced that can be linked with
atmospheric forcing as well as lateral freshwater sources. In fact it
is possible that the amount of cold water injected into the system
from the Po River, in the period preceding the DW event, and not
reproduced by the models, enters into the general circulation of the
basin and also affects the coastal area in front of Croatia
(Fig. , center). The comparison with the satellite image
from the 16 February shows the zone of highest bias (positive for SHYFEM
and negative for MITgcm) just south of the Po River delta, crossing
longitudinally the basin and in the transition zone between colder and
warmer waters. It seems that in this frontal zone, delimiting the area
with DW where vertical mixing would occur, the two models behave
differently. Another major discrepancy between the two models can be
seen in the narrow strip along the northernmost littoral, where SHYFEM
overestimates SST by around 1∘C and MITgcm
underestimates it by the same quantity. Generally SHYFEM has a bias in
the range [-0.5,+0.5]∘C in the offshore area of the NAS,
while MITgcm tends to overestimates SST by 1∘C there
(Fig. , bottom). To correctly interpret the outcomes from the
model–satellite comparison, we should highlight that there are several
factors,
which might affect the performance of SST satellite-derived results. Satellite-derived SST is the skin layer temperature and it provides information on only
a few microns of the sea surface. SST measured by buoys or derived by
models are generally collected at depths from 0.5 to 5 m below the sea surface.
These SSTs are called bulk SSTs.
Therefore, the skin SST can be significantly different from the bulk SST. Referring
to , surface thermal stratification can induce differences of
some degrees between the skin and the bulk temperatures. In the western Adriatic
Sea shelf, where the majority of river discharges occurs, the buoyancy flux due to
river runoff at the sea surface causes a significant increase in the difference
between the skin and the bulk temperatures. In addition, spatial variations in
the near-coast surface winds might induce different levels of heating in different
areas and generate spatial gradients in SST .
It has to be stressed that also water turbidity due to river runoff can affect the SST:
a modeling implementation in the Black Sea, done by and
demonstrated that high turbidity affects the depth corresponding to solar radiation
extinction and consequently the calculation of SST. demonstrated
that using a clear-water constant attenuation depth assumption (as done also in
the modeling work here proposed), as opposed to turbid water type values in the
modeling implementation, produced monthly SST biases as large as 2 ∘C in the
winter period in the Black Sea. Not being possible to apply different values of depth
corresponding to solar radiation extinction, based on the presence of sediments
(dynamics not simulated in the models), we had to take into account a possible
bias in simulating SST close to river inputs of 2∘C.
Time series of depth profiles of the (a) density anomaly,
(b) temperature, (c) salinity, and (d) RMS of
vorticity averaged over the north Adriatic area, for SHYFEM and
MITgcm, both in the hydrostatic and nonhydrostatic implementation.
Dense water formation and propagation
In Fig. we show, for SHYFEM and MITgcm (both in the
hydrostatic and nonhydrostatic implementation) time series of depth
profiles of the average density anomaly, temperature, salinity, and
root mean square (RMS) vorticity averaged over the NAS (in the shelf
region above latitude 44∘ N, for depths lower than
40m – area shown in Fig. ). The DW formation
event is marked by a strong increase in the density anomaly at the
beginning of February, with values reaching +5kgm-3
for SHYFEM, and slightly less for MITgcm. The SHYFEM values are in
agreement with those measured in . SHYFEM
describes the sudden formation of DW in the NAS and its sinking/mixing
over the whole water column (Fig. a). Also the subsequent
increase of the density anomaly at larger depths, during and after the
event, is detected by SHYFEM. From Fig. , the moment of DW
formation is clearly identified, for SHYFEM, after the first week of
February, lasting for 1 week and then the progressive decrease of
density anomaly marks its flow southward just out of the NAS. The
temperature profile time series (Fig. b) for SHYFEM
identifies the cold waters produced at the beginning of the
event. Interestingly, the bulk of cold water changes its
characteristics and temperature while sinking, even after the
event. This stresses the fact that DW characteristics are evolving,
being influenced by the mixing taking place with the surrounding
warmer waters. SHYFEM simulates an increase in surface salinity during
the DW event, suggesting evaporative processes due to the effect of
the cold Bora wind. Therefore, during the DW formation, SHYFEM has
a more homogeneous haline environment, highly thermally unstably
stratified, that leads to DW sinking. MITgcm, like SHYFEM also
registers the rapid increase in density anomaly at the beginning of
February, even if the rate of increase and the peak reached by MITgcm
is lower than SHYFEM. Similar temperature variations, with an unstable
stratification characteristic of the DW formation, are seen by both
models but it is less pronounced in SHYFEM. MITgcm has a lighter
water environment at the beginning of the simulation, compared with
SHYFEM, probably due to the presence of less saline waters on the
surface (Fig. c). This can be responsible for the lower
density anomaly simulated during the DW event by MITgcm. Another major
difference between the models is in the evolution of the bottom
salinity: MITgcm shows an increase just after the event, with an
higher stable haline stratification.
Maps of surface vorticity with surface current overlaid for SHYFEM
(left panel) and MITgcm (central panel), in the hydrostatic implementation,
and differences in vorticity between MITgcm hydrostatic and nonhydrostatic
implementations (HY-NH, right panel), for the dates indicated.
Maps of net vertical velocity with wind vectors overlaid for SHYFEM
(left panel) and MITgcm (central panel), in the hydrostatic implementation,
and differences in net vertical velocity between MITgcm hydrostatic and
nonhydrostatic implementations (HY-NH, right panel), for the dates indicated.
Red and blue colors in the net vertical velocity maps indicate upward and
downward motion.
From Fig. we can also compare the hydrostatic and
nonhydrostatic MITgcm simulations. For temperature, the nonhydrostatic
run stratifies thermally, just after the beginning of the simulation,
slightly more than the hydrostatic run. Also slightly colder waters
are present near the bottom, during and after the DW event, in the
nonhydrostatic MITgcm. This results in a relative increase of density
anomaly close to the bottom in February for the nonhydrostatic run,
though these small differences do not lead to any significant change
in the vertical dynamics between the two runs (Fig. d).
The DW formation corresponds with a strong increase in vorticity
detected by both models (Fig. d), though SHYFEM has a much
stronger and more prolonged vorticity intensification relative to the
MITgcm runs.
Circulation and vertical dynamics
In Figs. and , we show maps of surface vorticity
(surface currents overlaid) and net vertical velocity over the water column
(wind vectors overlaid), respectively, from SHYFEM and MITgcm for the 26
January, 5 and 16 February, and 27 March 2012. Additionally, maps of the
difference between MITgcm hydrostatic and nonhydrostatic runs are shown. The
26 January is characterized by spatially variable wind over the NAS. SHYFEM
has a narrow band of positive vorticity along the Italian littoral. Just off
the Po River delta, a band of negative vorticity is seen, probably due to the
advection of freshwaters out of the main branch (Fig. ).
A cyclonic circulation in front of the Venice Lagoon is detected by SHYFEM,
linked to upwelling there (Fig. ), directly induced by the wind.
MITgcm has the same patterns seen by SHYFEM but their values are about 1
order of magnitude lower, due to the much less energetic currents
(0.1ms-1 vs. 0.2ms-1 in SHYFEM). The direct
effect of wind forcing is seen in the surface vorticity map for the 5
February, during the DW formation event, both in SHYFEM and in MITgcm.
Clearly there is a strong enhancement of coastal upwelling along the eastern
coastline during the DW outbreak, as a result of the strong Bora winds
driving Ekman suction. In this case the vertical dynamics in SHYFEM is due to
the Ekman transport (coastal upwelling in the Gulf of Trieste coastal area)
as well as to the surface cooling by the Bora wind. Strong negative vertical
velocities indicating sinking are seen in the center of the Gulf of Trieste
and in the whole NAS coastal zone. Interestingly, as stated by
, other sources of DW are seen along the coast of
Croatia and in specific areas in the archipelago in front of it. MITgcm has
a general cyclonic circulation in the NAS, bordered by littoral negative
vorticity in the northern end of the basin. As with SHYFEM, the area of DW
sinking is seen but with a lower magnitude of vertical velocity, even if
a number of small-scale features are reproduced, showing higher horizontal
variability of vertical processes. On one hand, the higher resolution of
MITgcm over the NAS, allows for the reproduction of more small-scale vortical
structures, identifying a wider spatial range of processes, compared with
SHYFEM. On the other hand, the larger structures reproduced by SHYFEM (NAS
gyre during DW event) seem more energetic, with lower dissipation along the
vortices boundaries and lower large-to-small-scale energy turbulent cascade,
increasing the net vertical transport. The stronger horizontal surface
dynamics, registered in SHYFEM, can lead to energetic vortical structures
that enhance the larger-scale vertical dynamics connected to them.
Just after the DW formation, on the 16 February, Bora wind starts to
be weakened but is still present. SHYFEM shows a general surface
circulation, in the whole NAS, in the direction east–west, directly
following the wind curl. No specific downwelling is seen by SHYFEM,
while the coastal area of the Gulf of Trieste and the Croatia littoral
show upwelling, probably due to local effects of wind stress along the
basin border. MITgcm still has negative velocities in the NAS and
surface currents seem mainly directed along the north–south axis,
with meandering and small-scale patterns.
The low wind on the 27 March produces surface currents in the NAS with
different behaviors in the two models. SHYFEM shows weak but well-defined geostrophic circulation, going from east–west along the
coastline in the northern end of the basin. Coastal vertical movements
are not enhanced, except for a slightly positive vertical velocity in
the offshore areas on the NAS. MITgcm shows a counter current, in the
anticyclonic direction, with almost zero net vertical velocity.
Overall from Fig. , we see that the two models produce
different surface current patterns. SHYFEM has more energetic coastal
currents flowing southward, enhancing the freshwater transport out of
NAS. MITgcm, throughout the DW event, has weaker surface dynamics,
increasing the residence time of freshwaters in the NAS.
Figures and also provide insight into the role played by
nonhydrostatic processes. From the difference plots it is clear that
nonhydrostatic processes have no impact on the dynamics in the
shallowest coastal area of the NAS. Only in the deeper basin further
south do differences between the hydrostatic and nonhydrostatic
simulations appear, in particular along the slopes of the sills of the south
Adriatic.
To clarify how the surface forcings affect the two model simulations,
Fig. shows the total heat flux, in terms of gain/loss, for
the four dates presented above. As a general picture, for the dates
before and after the DW event, a lower gain of heat by MITgcm is seen,
compared with SHYFEM, while during the 5 February the heat loss is
more diffused for SHYFEM than for MITgcm. MITgcm has more local areas
of heat loss, directly connected with the Bora wind jets, but also shows
specific areas with small heat gain. In any case the values of
heat loss seen by the two models correspond with the ones also simulated
by .
Maps of surface net heat flux for SHYFEM (top) and MITgcm (bottom)
for the dates indicated. Units are [Wm-2].
(a) Time series of depth profiles of vertical velocity
averaged over the coastal area of the Gulf of Trieste, with depth lower than
18 m, and (b) time series of maximum daily Ekman wind curl and daily
maximum of net vertical velocity for SHYFEM and MITgcm.
In order to look closer at the coastal upwelling taking place, we examine the
vertical
velocity focusing on the coastal area of the Gulf of Trieste within
18m depth (Fig. ). SHYFEM and MITgcm time series
of vertical velocity profiles, averaged over this sub area are shown
(Fig. a). Here we present only the hydrostatic simulation
because no significant differences are seen in the nonhydrostatic run for
this area. A strong signal of positive (upward) velocity is detected by both
models during the DW event, though it is much stronger in SHYFEM than MITgcm.
This upwelling is the result of the net Ekman suction induced by the Bora wind
while there is a general DW sinking in the rest of the Gulf of Trieste. This
is evident in Fig. b, where we show the comparison between the
daily maximum Ekman induced vertical velocity (estimated from the two
different wind-stress formulations used in the two models) and the daily
maximum net vertical velocity. Both models show an Ekman induced upwelling
during the DW event, though the magnitude of the vertical velocity in
SHYFEM is more comparable to the Ekman values, whereas MITgcm shows
a lower net vertical velocity in the same period. The small differences
in the Ekman velocities computed by the two models are connected with
the different formulation of wind drag coefficients, though overall they
are very similar.
Maps of the depth at which the highest salinity gradient
(freshwater above, saline waters below) occurs, within the
37 psu isoline that identifies ROFI in the NAS, for the
dates indicated.
Riverine processes
Among the different coastal processes interacting during the DW event,
the riverine inputs have an important role to play that must be taken
into account. Different models can behave differently in reproducing
the river plumes shape, in terms of both horizontal spreading and
vertical mixing. Figure infers both of these aspects,
showing, for the four dates considered above, the depths where the
highest haline gradient (freshwater above saline water) occurs, within
the isoline of 37psu that is chosen as a limit to border
the ROFI environments. This choice mimics the approach proposed in
that identifies the plume limit at
36psu. We chose a slightly higher value, in order to
include the bulk zone of the plume and the relative mixed area in its
proximity. It has to be mentioned that, due to the low discharge
characterizing the simulated period that enhances the DW formation,
the ROFI is limited to a narrow coastal strip, except for the 27
March, when wind is weak and the discharges of rivers are relatively
higher than in the preceding period. As for the previous images, the
nonhydrostatic run is not shown due to the negligible differences in
the NAS, compared with the hydrostatic run.
There is always a wider extension of surface freshwater for the MITgcm
run than for SHYFEM. This is particularly evident along the northern
littoral of the basin and can be seen throughout the whole period
(Fig. ). Focusing on the major river in the area, the Po
River, it seems that the two models mixes differently along the water
column, with a higher freshwater stratification for MITgcm than for
SHYFEM, during periods of low wind and higher discharge (27
March). During the DW event, the effect of surface wind stress is high
and leads to a more confined strip of freshwater in both models,
though more so in SHYFEM, where the simulated stronger coastal current
is enhancing the freshwater flow southward along the littoral
(Fig. , 5 February). The major differences between the two
models are seen in the 16 February, when there is a much larger
surface spreading of freshwater in MITgcm.
Discussion
The reproduction of the majority of coastal processes, such as the DW
formation and its spreading southward, coastal upwelling, and estuarine
processes, requires taking into account a number of issues from
a modeling point of view. SHYFEM and MITgcm are two very different
models in terms of their numerical approach, parameterization, and
their treatment of boundaries and forcings.
The two models demonstrated major differences in reproducing the
correct amount of water with density anomaly higher than
5kgm-3 (Fig. ). The density anomaly
produced by SHYFEM during the DW event is higher and closer to the
measurements. The energy balances of the two models are different, as
can be deduced by the total heat maps shown in Fig. . Even
before considering how the dynamics acts on the water masses, it is
important that the correct energy is injected into the system that
will then be transferred into the vertical dynamics. The sinking
processes would lead to higher vertical velocities and initiate
stronger mixing with the surrounding waters due to the higher
thermohaline gradient.
The validation section revealed the importance of having the correct
setup in order to reproduce the predominant drivers of this
phenomenon, i.e., the mechanical action of wind, acting on the sea
surface, and the thermal flux due to the sudden cooling of the
air–sea interface. Thus, the availability of correct and adequately
resolved data set to force the models and a suitable treatment of
surface boundary stress and heat-mass fluxes are required. The two
models are forced with the same atmospheric data, but still have
different features on the surface: Fig. shows how SHYFEM
is able to capture values of density anomaly comparable with the ones
found by . This implies that the surface
forcings are realistic enough for the investigation of these phenomena
and the differences between the models' results can be linked
partially to the treatment of these forcings.
The two models apply different formulations in treating the wind
stress, inducing slightly different Ekman transports. However, despite
these differences, the Ekman velocities computed by the two models
have the same timing, particularly during the DW event
(Fig. b). On the local coastal scale, it seems that, even
if the formulation is different, the dynamics connected with wind
stress is similarly reproduced. Therefore, differences can be ascribed
more to other factors, namely the bulk formulas used to compute the
heat and mass surface fluxes.
In fact the choice of suitable bulk formulas that take into account
the specific processes connected with the air–sea interaction and the
heat and mass transfer through the interface strongly influence the
capability to reproduce the DW formation. The comparison both with the
Vida buoy and the Acqua Alta platform data shows different behaviors
by the two models, concerning the SST. The different choice adopted by
SHYFEM and MITgcm in dealing with the parameterization of the sensible
and latent heat fluxes give rise to the different results of the two
models, particularly after the DW event, approaching springtime when
there is an increase in the heat gained by the sea at the
surface. MITgcm has a warmer SST on the whole NAS area
(Figs. , and ). The slightly lower
heat flux through the surface computed by MITgcm (Fig. )
can result in a lower injection of energy and the lower dynamics seen
also in Figs. and . Figure also reveals that the
surface salinity in MITgcm during the DW formation is lower compared
with SHYFEM, which corresponds with the lower density anomaly
registered by the model. Different parameterizations of latent heat
adopted by the two models can play a role in this, in particular for
the computation of evaporation.
Moreover, the differences seen in the salinity are affected by the
lateral boundaries (i.e., river inputs), both in terms of forcing
availability and boundary treatment. Due to the lack of measured
discharge data for a number of rivers in the simulated area and the
use of river water temperature from 2007 at the Tagliamento River,
applied on all the lateral freshwater sources, the models show
discrepancies in matching the surface salinity temporal variability
(Figs. and ) and the spatial surface thermal
pattern close to river mouths (Fig. ). What is known is
that winter 2012, unlike the climatology, is colder than the
average. Therefore, the limit of both models, in providing the real
temperature of freshwater discharges in the NAS, and the use of
climatological discharge time series is responsible for the mismatch
found. Even if the use of the climatology is reasonable, at least for
what concerns the salinity, for a general characterization of the
dynamics as the model-measure biases are around 1psu, this
choice can deeply affect the reproduction of suddenly varying or
extreme events like the DW formation in winter 2012.
Moreover, even though the two models are forced with the same data sets
at the lateral boundaries, their surface biases have different signs,
suggesting that there is also a substantial difference in the momentum
laterally injected into the system and in the vertical mixing
simulated (Figs. and ). The two models apply
different approaches in dealing with lateral boundaries, with
differences in the reproduction of river mouth morphology, and in the
momentum applied for the freshwater discharge. The difference in
resolution, just along the coastline, that shows a higher-resolved
river channel shaping in SHYFEM, can affect the river discharge inflow
as a consequence of the geometry of the input points. The advection
induced in the transition zone between the narrow, well-defined river
channel and the open sea, as reproduced by SHYFEM, could be more
pronounced and could lead to different plume shaping. MITgcm includes,
as well, river channels but the lower resolution, just in the
proximity of estuaries and deltas, can affect the dynamics. Moreover,
MITgcm imposes velocity values at the upstream end of each channel,
and discharges are computed multiplying the values by the
cross sectional area. SHYFEM directly imposes the measured discharges
and momentum is injected into the system as a consequence of the water
level gradients between the boundary nodes and the surrounding
nodes. It is not possible to state if one of the two approaches is
more suitable, but as it has a possible effect, even if minor compared
with the geometric effect due to the different resolutions, it is worth
mentioning.
The different temperature and salinity fields simulated by the two
models close to the river mouths, mainly due to the freshwater
sources, provide different baroclinic gradients, affecting the coastal
thermohaline circulation. Additionally, the horizontal schemes used in
the models, for horizontal advection and diffusion of scalars
(i.e., temperature and salinity), would then lead to different
baroclinic currents and to the higher amount of freshwater in the sub
area of NAS for MITgcm, compared with SHYFEM (Fig. ). The
model biases computed at the Senigallia CTD transect
(Fig. ) are another example of this situation: coastal
haline stratification could be missed as a consequence of the Po River
plume mismatch, which is evident in the thermal bias shown in
Fig. that lead to the different riverine water spreading.
The model differences in the more offshore area of the NAS can also be
influenced by the different resolution of the two models: on one hand
it seems that higher resolution is needed along the coast, to
reproduce the complexity of the coastal morphologies; on the other
hand, a higher resolution offshore, as is the case for the MITgcm
grid, leads to a number of small-scale vortical structures, generally
missed by SHYFEM (Fig. ). It seems that SHYFEM is
horizontally more diffusive than MITgcm, but less dissipative (from an
energetic point of view). Therefore, less horizontal fronts are seen by
SHYFEM but they are steeper. An open point, that cannot be easily
discriminated from the obtained results, is the relative importance of
horizontal advection and mixing parameterization in the reproduction
of these processes. In order to add some information on the relative
effect of horizontal viscosity parameterization, a preliminary test
simulation with MITgcm was carried out, though not shown here. The
variable horizontal viscosity parameterization, proposed by
and already implemented in for
the Gibraltar Strait, was checked. Enhanced vorticity patterns,
compared with the hydrostatic MITgcm run discussed in the paper were
seen, suggesting the horizontal viscosity plays a role in correctly
reproducing horizontal features. Consequently, effects on the vertical
transports were detected. Further investigation of this topic is left
for future work.
Analyzing the models' capability to reproduce the vertical
hydrodynamic field structure, as a general comment, it seems that
SHYFEM outputs are characterized by higher vertical mixing compared
with MITgcm, that can be ascribed to both the higher thermal energy
gradient between water masses simulated in the process and the
vertical mixing connected with the turbulence closure schemes used.
Finally, Figs. and suggest the neglectable importance of
nonhydrostatic processes in producing the dense water formation and
coastal upwelling in the NAS. These processes are governed by the mass
balance, in terms of horizontal transfer of water due to wind
vs. vertical suction of water to compensate, which are already
reproduced with the hydrostatic approximation.
Conclusions
The coastal zone of the NAS is characterized by a number of hydrodynamic processes
that interact and evolve on different spatial and temporal scales. The present
work demonstrates the complexity of modeling these specific processes and
identifies a number of issues needed for choosing the most suitable modeling
strategy for this typology of study. The main findings are listed below.
The two models use different bulk formulas for surface latent and
sensible heat, accordingly to their state-of-the-art setups already tested
in the area. This leads to different heat transfer at the surface, giving rise
to an overall different energy balance. Lower convective dynamics over the
water column is reproduced in the case of MITgcm relative to SHYFEM.
Therefore, the choice of suitable bulk formulas, specifically in the coastal zone,
is a central point for modeling implementations.
There are differences in the small-scale hydrodynamic structure
in the offshore area of NAS that are connected with higher resolution over
the whole domain in MITgcm. However, these fine-scale features in MITgcm
have little impact on the overall reproduction of the dense water formation;
therefore, the presented implementation identifies the spatially variable minimum
resolution adequate to reproduce the investigated processes, that span from
less than 500 m in the nearshore area up to 1–2 km offshore. Higher
resolutions do not add information on the main investigated dynamics.
A highly resolved coastal zone, with the possibility to
reproduce the complex morphology connected with lateral freshwater
inputs, can provide the correct momentum injection into the system
and affects the capability to reproduce buoyant processes in the
coastal area.
Nonhydrostatic processes have little impact on the coastal
features seen on the shelf of the NAS, suggesting that the hydrostatic
models are adequate for simulating DW formation in the shallow areas of the basin.
There are a number of outstanding issues that are not tackled in this present work.
It is important to point out that in this paper we used different wind stress formulation
and bulk formulas for the two models, which result in uncertainties in the model comparison.
Further studies, perhaps using direct forcing with heat and mass fluxes provided by
meteorological models, would be helpful in reducing these uncertainties. Other open
questions not considered in this work are the effects that different horizontal advection,
mixing, and turbulence closure schemes have on coastal hydrodynamic processes, such as
the dense water event considered here. Such a study would require using a single model
with different implementations of these schemes to precisely characterize and attribute their
impact on the coastal dynamics. Also, here we found that nonhydrostatic processes have
little impact in the shallow coastal shelf of the NAS, though there were differences from the
hydrostatic case seen in the deeper part of the basin. Exploring the entire Adriatic basin
may reveal if the nonhydrostatic dynamics plays any part in the wider propagation of dense
water through the basin, particularly in the southern Adriatic pit.
Despite these outstanding questions, this work provides some clarity on the chosen setups
that were already used for implementations in the Adriatic Sea, giving suitable suggestions
for improvements. These modeling implementations, mainly devoted to process investigations,
can be used to guide choices made for possible future operational products.
Acknowledgements
This work is funded by RITMARE Flagship Project – National Research
Programme (Italian Ministry of University and Research). Data from
Vida buoy were provided by the National Institute of Biology, Marine
Biology Station, Piran, Slovenia, thanks to Vlado Malacic. Thanks to
Alfredo Boldrin (ISMAR-CNR, Venezia) and the Veneto Region (Regional
Law no. 15, 12 July 2007, Projects SISOE: Integrated System for
Oceanographic and Ecological monitoring of the Adriatic Sea and
MARINA), for the temperature and salinity data at the CNR Platform
Acqua Alta. Thanks to Federica Grilli and Mauro Marini (ISMAR-CNR,
Ancona) for the Senigallia CTD transects.
The computing resources and the related technical support used for running
the MITgcm model have been provided by CRESCO/ENEAGRID High Performance
Computing infrastructure and its staff . CRESCO/ENEAGRID High
Performance Computing infrastructure is funded by ENEA, the Italian National
Agency for New Technologies, Energy and Sustainable Economic Development
and by Italian and European research programmes, see
http://www.cresco.enea.it/english for Information.
Edited by: L. Kantha
References
Bellafiore, D. and Umgiesser, G.: Hydrodynamic coastal processes in
the North Adriatic investigated with a 3D finite element model,
Ocean Dynam., 60, 255–273, 2010.
Benetazzo, A.,
Bergamasco, A., Bonaldo, D., Falcieri, F. M., Sclavo, M.,
Langone, L., and Carniel, S.: Response of the Adriatic Sea to an
intense cold air outbreak: dense water dynamics and wave-induced
transport, Prog. Oceanogr., 128, 115–138, 2014.
Bougeault, P. and Lacarrere, P.: Parametrization of
orography-induced turbulence in a mesoscale model, Mon. Weather
Rev., 117, 1872–1890, 1989.
Burchard, H. and
Peterson, O.: Models of turbulence in the marine environment –
a comparative study of two equation turbulence models, J. Marine
Syst., 21, 29–53, 1999.
Burrage, D. M.,
Book, P. J., and Martin, P. J.: Eddies and filaments of the Western
Adriatic Current near Cape Gargano: analysis and
prediction, J. Marine Syst., 78, S205–S226, 2009.
Clark, N. E., Eber, L.,
Laurs, R. M., Renner, J. A., and Saur, J. F. T.: Heat exchange
between ocean and atmosphere in the Eastern North Pacific for
1961–1971, Tech. Rep. NMFS SSRF-682, NOAA, US Department of Commerce,
Washington, D.C., USA, 1974.
Cushman-Roisin, B., Gačić, M., Poulain, P.-M., and
Artegiani, A.: Physical Oceanography of the Adriatic Sea, Springer, the
Netherlands, 304 pp., 2001.
Drofa, O. V. and Malguzzi, P.: Parameterization of microphysical processes in
a non hydrostatic prediction model,
Proceeding in 14th International Conference on Clouds and Precipitation (ICCP), Bologna,
19–23 July 2004, 1297–3000, 2004.Doglioli, A., Griffa, A., and Magaldi, M. G.: Numerical study of a coastal current
on a steep slope in presence of a cape: the case of the Promontorio
di Portofino, J. Geophys. Res., 109, C12033, 10.1029/2004JC002422, 2004.
Donlon, C. J., Minnett, P. J., Gentemann, C., Nightingale, T. J., Barton,
I. J., Ward, B., and Murray, M. J.:
Toward improved validation of satellite sea surface skin temperature measurements
for climate research, J. Climate, 15, 353–369, 2002.Durrieu de Madron, X., Houpert, L., Puig, P., Sanchez-Vidal, A.,
Testor, P., Bosse, A., Estournel, C., Somot, S., Bourrin, F.,
Bouin, M. N., Beauverger, M., Beguery, L., Calafat, A., Canals, M.,
Cassou, C., Coppola, L., Dausse, D., Ortenzio, F. D., Font, J.,
Heussner, S., Kunesch, S., Lefevre, D., Le Goff, H.,
Martín, J., Mortier, L., Palanques, A., and Raimbault, P.:
Interaction of dense shelf water cascading and open-sea convection in the
northwestern Mediterranean during winter 2012, Geophys. Res. Lett., 40,
1379–1385, 10.1002/grl.50331, 2013.Durski, S. M., Glenn, S. M., and Haidvogel, D.: Vertical mixing schemes in
the coastal ocean: comparision of the level 2.5 Mellor-Yamada scheme with an
enhanced version of the K profile parameterization, J. Geophys. Res., 109,
C01015, 10.1029/2002JC001702, 2004.
Falcieri, F. M.,
Benetazzo, A., Sclavo, M., Russo, A., and Carniel, S.: Po River
plume pattern variability investigated from model data,
Cont. Shelf Res., 87, 84–95, 2014.Ferrarin, C., Roland, A., Bajo, M., Umgiesser, G., Cucco, A., Davolio, S.,
Buzzi, A., Malguzzi, P., and Drofa, O.: Tide-surge-wave modelling and
forecasting in the Mediterranean Sea with focus on the Italian coast, Ocean
Model., 61, 38–48, 10.1016/j.ocemod.2012.10.003, 2013.
Garvine, R. W.: A dynamical system for classifying buoyant coastal discharges,
Cont. Shelf Res., 15, 1585–1596, 1995.
Gaspar, P., Gregoris, Y.,
and Lefevre, J.-M.: A simple eddy kinetic energy model for
simulations of the oceanic vertical mixing: tests at station papa
and long-term upper ocean study site, J. Geophys. Res., 95,
16179–16193, 1990.
Janekovic, I.,
Mihanović, H., Vilibić, I., and Tudor, M.: Extreme cooling and
dense water formation estimates in open and coastal regions of the
Adriatic Sea during the winter of 2012, J. Geophys. Res., 119,
3200–3218, 2014.
Jiang, L., Yan, X.-H.,
Tseng, Y.-H., and Breaker, L. C.: A numerical study on the role of
wind forcing, bottom topography, and nonhydrostacy in coastal
upwelling, Estuar. Coast. Shelf S., 95, 99–109, 2011.
Kara, A. B., Hurlburt, H. E., Rochford, P. A., and O'Brien, J. J.:
The Impact of Water Turbidity on Interannual Sea Surface Temperature
Simulations in a Layered Global Ocean Model, J. Phys. Oceanogr.,
34, 345–359, 2004.
Kara, A. B., Wallcraft, A. J., and Hurlburt, H. E.:
Sea Surface Temperature Sensitivity to Water Turbidity from
Simulations of the Turbid Black Sea Using HYCOM, J. Phys. Oceanogr.,
35, 33–54, 2005.
Kondo, J.: Air–sea bulk transfer coefficients in diabatic
conditions, Bound.-Lay. Meteorol., 9, 91–112, 1975.
Large, W. and Pond, S.:
Open ocean momentum flux measurements in moderate to strong
winds, J. Phys. Oceanogr., 11, 324–336, 1981.
Large, W. and Pond, S.:
Sensible and latent heat flux measurements over the
ocean, J. Phys. Oceanogr., 12, 464–482, 1982.
Leith, C.: Parameterization of
vertical mixing in numerical-models of tropical oceans,
Phys. Fluids, 10, 1409–1416, 1968.Magaldi, M. G. and
Haine, T. W. N.: Hydrostatic and non-hydrostatic simulations of
dense waters cascading off a shelf: the East Greenland case,
Deep-Sea Res. Pt. I, 96, 89–104, 10.1016/j.dsr.2014.10.008, 2015.
Magaldi, M. G.,
Özgökmen, T. M., Griffa, A., and Rixen, M.: On the response
of a turbulent coastal buoyant current to wind events: the case of
the Western Adriatic Current, Ocean Dynam., 60, 93–122,
2010.
Mahadevan, A.: Modeling
vertical motion at ocean fronts: are nonhydrostatic effects relevant
at submesoscales?, Ocean Model., 14, 222–240, 2006.Malguzzi, P., Grossi, G., Buzzi, A., Ranzi, R., and Buizza, R.: The 1966
“century” flood in Italy: A meteorological
and hydrological revisitation, J. Geophys. Res., 111, D24106,
10.1029/2006JD007111, 2006.
Marshall, J.,
Hill, C., Perelman, L., and Adcroft, A.: Hydrostatic,
quasi-hydrostatic, and nonhydrostatic ocean
modeling, J. Geophys. Res., 102, 5733–5752, 1997a.
Marshall, J.,
Adcroft, A., Hill, C., Perelman, L., and Heisey, C.:
A finite-volume, incompressible Navier Stokes model for studies of
the ocean on parallel computers, J. Geophys. Res., 102, 5753–5766,
1997b.Mihanović, H.,
Vilibić, I., Carniel, S., Tudor, M., Russo, A., Bergamasco, A.,
Bubić, N., Ljubešić, Z., Viličić, D., Boldrin, A.,
Malačič, V., Celio, M., Comici, C., and Raicich, F: Exceptional
dense water formation on the Adriatic shelf in the winter of 2012, Ocean
Sci., 9, 561–572, 10.5194/os-9-561-2013, 2013.
Moum, J. N., Nash, J. D., and
Klyman, J. M.: Small-scale processes in the coastal ocean,
Oceanography, 21, 22–33, 2008.
Orlić, M. and
Pasarić, Z.: A simple analytical model of periodic coastal
upwelling, J. Phys. Oceanogr., 41, 1271–1276, 2011.Orlić, M., Paklar, G. B., Dadić, V., Leder, N., Mihanović, H.,
Pasarić, M., and Pasarić, Z.: Diurnal upwelling resonantly
driven by sea breezes around an Adriatic island, J. Geophys. Res.,
116, C09025, 10.1029/2011JC006955, 2011.Otero, P., Ruiz-Villarreal, M., and Peliz, Á: River plume fronts off
NW Iberia from satellite observations and model data,
ICES J. Mar. Sci., 66, 10.1093/icesjms/fsp156, 2009.Ponti, G., Palombi, F., Abate, D., Ambrosino, F., Aprea, G.,
Bastianelli, T., Beone, F., Bertini, R., Bracco, G., Caporicci,
M., Calosso, B., Chinnici, M., Colavincenzo, A., Cucurullo,
A., Dangelo, P., De Rosa, M., De Michele, P., Funel, A., Furini,
G., Giammattei, D., Giusepponi, S., Guadagni, R., Guarnieri, G.,
Italiano, A., Magagnino, S., Mariano, A., Mencuccini, G., Mercuri,
C., Migliori, S., Ornelli, P., Pecoraro, S., Perozziello, A., Pierattini,
S., Podda, S., Poggi, F., Quintiliani, A., Rocchi, A., Scio, C., Simoni,
F., and Vita, A.:
The role of medium size facilities in the HPC ecosystem: the case of the new
CRESCO4 cluster integrated in the ENEAGRID infrastructure, Proceedings of
the 2014 International Conference on High Performance Computing and Simulation,
HPCS 2014, art. no. 6903807, 1030–1033, 10.1109/HPCSim.2014.6903807, 2014.
Querin, S., Crise, A.,
Deponte, D., and Solidoro, C.: Numerical study of the role of wind
forcing and freshwater buoyancy input on the circulation in
a shallow embayment (Gulf of Trieste, Northern Adriatic
Sea), J. Geophys. Res., 111, C03S16,
10.1029/2006JC003611,
2006.
Querin, S., Cossarini, G.,
and Solidoro, C.: Simulating the formation and fate of dense water
in a midlatitude marginal sea during normal and warm winter
conditions, J. Geophys. Res.-Oceans, 118, 885–900,
2013.
Reinart, A. and Reinhold, M.: Mapping surface temperature in large lakes with
MODIS
data, Remote Sens. Environ., 112, 603–611, 2008.
Russo, A. and Artegiani, A.: Adriatic sea hydrography, Sci. Mar., 60, 33–43, 1996.
Sanchez-Garrido, J.,
Sannino, G., Liberti, L., and Garcia Lafuente, J. L. P.: Numerical
modeling of three-dimensional stratified tidal flow over camarinal
sill, Strait of Gibraltar, J. Geophys. Res., 116, 1978–2012,
2011.Sannino, G., Sanchez
Garrido, J. C., Liberti, L., and Pratt, L.: Exchange flow through
the Strait of Gibraltar as simulated by a-coordinate hydrostatic
model and a z-coordinate nonhydrostatic model, in: The
Mediterranean Sea: Temporal Variability and Spatial Patterns, John
Wiley & Sons Inc., Oxford, UK, 25–50, 2014.
Thomas, L. N., Tandon, A., and
Mahadevan, A.: Submesoscale processes and dynamics, in: Ocean
Modeling in an Eddying Regime, Geoph. Monog. Series, 177,
17–38, 2008.
Umgiesser, G.,
Canu, D. M., Cucco, A., and Solidoro, C.: A finite element model for
the Venice Lagoon. Development, set up, calibration and
validation, J. Marine Syst., 51, 123–145, 2004.
Vested, H. J.,
Berg, P., and Uhrenholdt, T.: Dense water formation in the Northern
Adriatic, J. Marine Syst., 18, 135–160, 1998.
Vilibić, I. and Mihanović, H.: Observing the bottom density
current over a shelf using an Argo profiling float, Geophys. Res. Lett., 40, 910–915, 2013.
Vilibić, I. and
Supić, N.: Dense water generation on a shelf: the case of the
Adriatic Sea, Ocean Dynam., 55, 403–415, 2005.
Wolanski, E., Richmond, R., Davis, G., Deleersnijder, E., and Leben, R.:
Eddies
around Guam, an island in the Mariana Islands group,
Cont. Shelf. Res., 23, 991–1003, 2003.